LIBS spectrum and image information fusion for waste aluminum alloy classification

Traditional waste aluminum (Al) alloy sorting technologies face technical bottlenecks such as low efficiency and poor accuracy. Image-based classification methods struggle too accurately identify alloy compositions and grades due to the similar visual characteristics of surface textures of different Al alloys. Laser-induced breakdown spectroscopy (LIBS), as an emerging spectral analysis technique, offers the capability for rapid chemical composition analysis of samples. However, signal fluctuations and the spectral line interference caused by the matrix effects and heterogeneity constrain the analytical accuracy and stability of LIBS technology. Therefore, this paper proposes an improved ResNet18-SVM fusion classification model, by leveraging image-spectral bimodal data, to improve accuracy and robustness. It extracts image features via an improved lightweight ResNet18 network, combines them with LIBS spectral features processed by SVM, and employs a Random Forest classifier for final decision-making, achieves efficient and accurate classification. Experimental results demonstrate that the proposed model exceeds 97% on accuracy and recall metrics. It is significantly superior to the ResNet18 model based on pure images and the SVM method based on pure spectra. Furthermore, it achieves an 72% reduction in classification time compared to the original ResNet18 model, demonstrating high precision and efficiency. This model provides an effective solution for waste Al alloy intelligent sorting, holds broad application prospects.